Inside combat on Cloud Mythos 5

While the rest of the country celebrated the United States’ first World Cup victory and the New York Knicks’ championship, Anthropic spent the weekend battling the Trump administration over its latest release. At 5:21pm on Friday, the company received a US export control directive to suspend access to the Mythos 5 and Fable 5 AI models by “any foreign national” inside or outside the United States, “including foreign national anthropic employees.” The only way that was possible, the Anthropist decided, was to do it completely revocation Products she spent The hype last week -And traveling to Washington, D.C., in hopes of changing President Donald Trump’s mind. Now, over the coming days, the US government could dramatically change the course of the entire industry, dealing a heavy blow to US AI companies.

Claude Mythos 5 and Fable 5 are built on the same foundation as Anthropic’s Mythos Preview, which It’s called a very dangerous anthropist To be released publicly. (The company’s warnings could be viewed as a real concern or more hype for its own model — or both.) Mythos 5 was made available to a select group of government agencies and companies, while Fable 5, which included additional safeguards, was deemed “safe for general use.” But when a report suggests that these guardrails may have failed, Anthropic’s dire warnings about Mythos falling into the wrong hands come back to haunt her.

A source familiar with the situation, who was involved in negotiations between Anthropic and the Trump administration, said the administration contacted the AI ​​lab on Friday around 1 p.m. ET and gave the company a 90-minute ultimatum to close access to Mythos 5 and Fable 5. If that doesn’t happen, the government will impose export controls on Anthropic under the authority of the U.S. Department of Commerce.

Anthropic executives were speaking to the White House within 15 minutes of that first call, the source said, confirming that CEO Dario Amodei joined the discussions about an hour and 15 minutes after that initial call. The source confirmed that Amodei spoke directly with US Treasury Secretary Scott Besent, Commerce Secretary Howard Lutnick, and National Internet Director Sean Cairncross, and in some cases more than once.

Anthropic wrote in a statement on Friday that the company believes the government “believes it has become aware of a method to bypass or ‘jailbreak’ Fable 5.” However, rather than an existential threat, Anthropic said the jailbreak in question was a “potentially narrow and non-global” operation that was “shared with the government” by an entity the company declined to name. Furthermore, Anthropic said the behavior was not unique to Fable 5. “We reviewed a report that we believe is the basis of government guidance and confirmed that the level of capability demonstrated there is broadly available from other models (including OpenAI’s GPT-5.5),” Anthropic wrote.

traffic lights I mentionedCiting an informed source, the uproar began because the US government was concerned about a China-linked group gaining access to the technology. But the source said the China-related rumors go back weeks, referring to a large global telecommunications company that was initially allowed to be included in Mythos Preview access, and that when the US government shared its concerns, Anthropic immediately revoked access.

that Share X By David Sachs, the former US government AI and cryptocurrency official who resigned in March, also did not mention China. However, Sachs did mention the unnamed entity that disclosed the problem to the government, calling it “a very trusted partner of both Anthropic and USG that was testing Fable (which) came forward with a jailbreak for those barriers.”

some Reports Point to Amazon CEO Andy Jassy as the person who reported his concerns to the US government after Amazon researchers red teamed Fable 5. This conclusion is at odds with some independent red team members, who have He said They were impressed by the level of protection.

The source familiar with the negotiations said Amazon’s research was explicitly mentioned in talks with the US government. Anthropic gained access to that paper within days of Friday’s export control directive, and has been going back and forth ever since with Amazon researchers to discuss it, the person added.

Everything in this paper can be achieved by OpenAI’s GPT-5.5, the source said.

Anthropic spent the weekend scrambling to play nice with the Trump administration, starting with virtual meetings and then flying employees to D.C., including Dave Orr, Anthropic’s head of collateral; Logan Graham, who runs the Red Frontier Team and led work on the Glasswing project; and Nicholas Carlini, a leading developer and cybersecurity researcher. Axios I mentionedCiting a source familiar with the Trump administration’s thinking, the company simply made repeated mistakes in its communications with the administration and that it “did not do a great job of trying to talk to the administration and appreciate ideological differences.” For Anthropic, the timing couldn’t be worse: the company relied on Mythos to do just that Help him recoverIn part, the result of months of high-level clashes with the US Department of Defense.

The source familiar with the negotiations said Anthropic had previously briefed management on Fable 5, and that the US Commerce Department had conducted pre-deployment testing, without sharing any concerns at the time. The source added that Anthropic has been working closely with government agencies since the release of the Mythos Preview.

The Trump administration initially took a hands-off approach to AI safety, but after the myths, it did so You become more contradictoryeven as it worries about the threat of losing the AI ​​race to China. Now, prominent cybersecurity leaders are as well to caution Sidelining Mythos 5 and Fable 5 could give China a huge advantage in the field of AI. Trump’s move has happened Galvanized international calls To search for alternatives to American artificial intelligence systems, while putting the new main model of a major American company in the field of artificial intelligence on ice.

A Public speech Technology and cybersecurity executives called for Fable 5 to be rescinded on Sunday. “Not all of us agree that regulation of AI is the right way forward,” the letter said, adding that if regulation is implemented regardless, it should be rooted in “scientific assessments developed with input from industry and academia.”

said Alex Stamos, Chief Product Officer at Corridor Edge He organized the public letter because the countless vulnerabilities of the past decade-plus, written in a variety of different programming languages, require AI patching before bad actors find them. “We’re in a race, and I think policymakers don’t understand that,” Stamos said. “There’s this weird arrogance, this idea that American labs are vastly superior to our opponents which will always be true, and that it’s really important to restrict access because of that. I think that’s stupid. If the labs were ahead, it would only take a few months. And you can see that in the open evaluations. The cutting-edge models are only about six months ahead of the Chinese models — and those are the models we know.”

The public letter goes on to note that while Anthropic’s Mythos class models are skilled at finding cybersecurity vulnerabilities and leveraging exploits, they are not “uniquely good” at these tasks and that Fable 5’s safeguards “were so aggressive that they were the source of humor in the cyber community on launch day.” Stamos said Edge that “There’s been a real overestimation of Mythos’ abilities. And Anthropic is clearly somewhat responsible for that herself… Mythos is great, but the real turning point was last year.”

Stamos said the industry is rife with standby contracts being signed with non-U.S. companies and open-weight models being deployed on standby hardware arrangements because the past weekend made political risk a part of companies’ business plans more than ever before.

“They’re laughing at us in Beijing right now,” Stamos said. “One of America’s heroes was beaten by the US government while we were in a race with the Chinese. It’s so stupid. That’s why I wrote the letter, and I think that’s why so many people signed it.”

said Ben Van Roo, co-founder and CEO of Legion Intelligence, an operative system for the national security community Edge “The directive that no foreign national should use this form is impossible to enforce.” “When I first read that, my entire network of geeks in the AI ​​community was having a blast,” he added.

To make matters more pressing, OpenAI, Google, and Microsoft have come out with their own products similar to Anthropic’s Mythos, making many of the same claims about its effectiveness and risks. If the Trump administration bans Anthropic’s advanced cybersecurity models, it could make a case for banning its competitors’ models as well. This could motivate AI industry leaders to unite and help Anthropic, or, as with its battle with the Pentagon over autonomous weapons, position themselves as a powerful force. A safer and more compliant alternative.

Even as the Trump administration tries to do so Free technology companies Given the regulatory hassles, the humanitarian system could amount to a significant restriction on powerful AI models, depending on how things play out in the next few days.

Legion Intelligence’s Van Roo described it as “uncharted territory” in the regulatory framework, adding that he did not believe this was the last time something like this would happen.

We have also entered an era of AI populism, as a growing number of people resist the AI ​​industry’s outsized influence and the concentration of power at the top through data center protests, pledges to… Leaves With AI chatbots, wrongful death lawsuits, and even attempts to attack AI company CEOs. The Trump administration’s recent moves against Anthropics could raise “greater fears and anxieties, perhaps for the wrong reasons,” Van Roo says.

The source familiar with the negotiations described the weekend talks as constructive, with some members of the administration acknowledging that imposing export controls on model providers is not ideal, because competitors with similar products could find themselves under the same restrictions — and since the U.S. government Explore offline A program that would encourage the export of American artificial intelligence systems.

As Anthropic continues to negotiate with the US government, there is little chance that the company’s myriad other issues will not come up with the Pentagon – the ongoing battle between Anthropic and the Department of Defense over policies for the acceptable use of Anthropic’s technology by the US military.

“This is new and we haven’t encountered anything that potentially radical before, and it has some real ramifications” in terms of how to enforce access to robust models, Van Roo said. “Who can use this new technology that still exceeds our ability to regulate it?”